To the Editor:

Recent studies have identified mutational genomic signatures introduced by Apolipoprotein B mRNA-Editing Catalytic Polypeptide-like (APOBEC) deaminases as well as inflammatory processes as being pivotal for MM onset and progression [1,2,3,4]. Although these recent insights provide a better understanding of MM pathogenesis, they have not yet been translated into clinical applications such as MM risk stratification. The current standards for MM patient risk classification are the International Staging System (ISS), the Revised ISS (R-ISS) and the second revision of the R-ISS (R2-ISS) introduced between 2005 and 2022, respectively [5]. All scores are based on clinical parameters reflecting tumor burden, and the newer R-ISS and R2-ISS further incorporate high-risk cytogenetics [5]. Considering that most risk-defining chromosomal abnormalities reflect early events in MM cells [6], we concluded that tumor burden and/or cytogenetics-based classifiers might not accurately reflect the dynamics of disease progression in MM patients. Therefore, we hypothesized that a predictive score which reflects molecular mechanisms that drive MM progression, can improve the accuracy of current MM risk classifiers. To test this hypothesis, we constructed and validated a proof-of-principle risk classifier called Editor/Inflammation- or EI-score, which combines mRNA levels of survival-associated APOBEC genes, pro/anti-inflammatory genes as well as clinical markers for MM disease burden.

Data from 1143 patients with newly diagnosed MM (NDMM) and available survival information was obtained through the CoMMpass database version IA14, which was generated as part of the Multiple Myeloma Research Foundation (MMRF) Personalized Medicine Initiatives (www.themmrf.org). ISS, R-ISS and R2-ISS staging information was available for 1113, 694, and 694 patients, respectively. For 599 patients, information on both blood parameters and RNA-seq was available. As an independent validation cohort, we analyzed clinical, cytogenetic, and RNA-seq data from 263 NDMM patients treated as part of the IFM/DFCI 2009 trial (ClinicalTrials.gov identifier: NCT01191060) [7]. IFM/DFCI patients were treated with Bortezomib, Lenalidomide and Dexamethasone (VRD) alone or with VRD+autologous stem cell transplantation (ASCT). All patient baseline characteristics (CoMMpass and IFM/DFCI) are summarized in Table S1. A stepwise workflow for the evaluation and selection of individual features and multivariate models in the MMRF CoMMpass dataset is shown in Fig. S1 and described in detail in the Supplementary Methods.

To translate recent whole genome- and RNA-sequencing findings into a predictive score, we pre-selected 163 features, including demographic, clinical, genomic, and cytogenetic information, as well as inflammatory signaling and nucleotide editing-associated mRNA covariates from the MMRF CoMMpass dataset (Fig. S1). Of the 163 tested variables, 25 for overall survival (OS) and 21 for progression-free survival (PFS) showed significant time-to-event outcomes. Notably, only one out of five cytogenetic features, namely +1q/amp1q (Fig. S2), passed our stringent selection criteria in 599 NDMM patients. In line with our hypothesis, we found that mRNA levels of individual APOBEC genes as well as APOBEC-induced genomic mutational signatures (calculated in form of both COSMIC single-base substitution (SBS) signature and APOBEC mutation enrichment score [8, 9]) were associated with inferior OS and PFS (Fig. S2). As the rationale of this study was not to provide a score for immediate clinical application but rather to determine if combining APOBEC and inflammation-associated gene expression variables holds prognostic merit for MM patients, we reduced our feature set to only the most significant variables that were associated with both OS and PFS. We then combined all age- and treatment-independent prognostic variables that passed our selection criteria (and for RNA parameters, showed a median expression >5 fragments per kilobase per million) into multivariate CoxPH models, excluding patient cytogenetics and mutational signatures. This included the following parameters: ß2M, Creatinine, Hemoglobin, LDH, APOBEC2, APOBEC3A, APOBEC3B, APOBEC3C, APOBEC3D, APOBEC3F, APOBEC3G, IL10, IL11, IL17C, IL27, IFNG, TGFB1, TGFB3, IL22RA1, IL2RA, TGFBR3, CXCL13. Patient age >75 y was excluded due to the inclusion criteria of the IFM/DFCI2009 study (18–65 y). The multivariate model with the highest predictive performance while retaining as few parameters as possible included the following seven features: ß2M, LDH, APOBEC2, APOBEC3B, IL11, TGFB1, TGFB3. Based on these seven parameters, we devised a streamlined scoring formula that relies on maximally selected rank statistics established cut-offs and incorporates weights derived from the rounded integer multivariate CoxPH z-score of each parameter. Although we detected strong correlation among expression levels of most members of the APOBEC family, there was no significant positive correlation between APOBEC2 and APOBEC3B (Pearson’s R = 0.039), which are both part of the EI-score (Fig. S3). The distribution of each expressed EI-score gene in the different MMRF CoMMpass cytogenetic and age groups is shown in Fig. S4.

To evaluate the prognostic accuracy of the EI-score compared to ISS, R-ISS, R2ISS, and mSMARTcyto (a reduced version of the Mayo clinic mSMART score: https://www.msmart.org, based on the presence of t(4;14), t(14;16), t(14;20), +1q and/or del(17p)), we computed performance metrics for the outcome prediction of each score in MMRF CoMMpass patients (Table 1, Fig. S5). The EI-score achieved the best performance for OS and PFS prediction (n = 599; Concordance index (Ci) 0.7 and 0.69, respectively), followed by R2-ISS (n = 694; Ci 0.66 and 0.61), ISS (n = 1113; Ci 0.66 and 0.6), R-ISS (n = 690; Ci 0.64 and 0.6), and mSMARTcyto (n = 823; Ci 0.58 and 0.54). We then successfully validated the EI-score in the IFM/DFCI2009 NDMM cohort (n = 263) (Fig. S5), representing a homogeneously treated patient collective. Notably, addition of EI-score gene expression information to ISS, R-ISS, R2-ISS (Fig. 1A, Table 1, Table S2), and mSMARTcyto, improved the performance of each classifier significantly. Moreover, applying the EI-score exclusively to MM patient subgroups with del(17p), +1q, and t(4;14) allowed to identify previously unrecognized favorable risk patients with adverse risk cytogenetics in the MMRF CoMMpass (Fig. 1B, Fig. S6) as well as in the IFM/DFCI cohort (Fig. 1C). In line, we found that del(17p), +1q, and t(4;14) patients with a high EI-score, displayed an enrichment of APOBEC-induced genomic mutations compared to low/intermediate EI-score patients (Fig. S7). These results demonstrate that the integration of APOBEC and inflammatory cytokine mRNA levels improve the prognostic capacity of chromosomal abnormalities, which are currently viewed as risk class defining. To adjust for the heterogeneous treatment protocols of patients included in the MMRF CoMMpass dataset, we also conducted a sub-analysis of MM patients receiving Cyclophosphamide, Bortezomib, Dexamethasone (CyBorD) or VRD ± ASCT (Fig. S8) and a sub-analysis of MM patients receiving VRD ± ASCT + maintenance therapy (Fig. S9), in which the EI-score also outperformed ISS, R-ISS, and R2-ISS. A possible explanation why APOBEC family members have so far not been part of probe-based mRNA classifiers such as EMC-92 [10] and UAMS-70 [11] is likely due to their high sequence similarity resulting in probe cross-hybridization and multimapping to several APOBEC members [12]. The high hazard ratio and predictive performance of APOBEC3B expression for adverse PFS and OS which appears to be independent from that of APOBEC-induced mutational signatures, likely reflects APOBEC3B’s additional involvement in MM pathogenesis through immune editing, viral and retroelement restriction, DNA demethylation, and tissue homeostasis [13]. Although APOBEC3B-induced C-to-U lesions are typically resolved by DNA repair response mechanisms, they can promote chronic replication stress and thus contribute to MM development, which could be a reason for the high predictive value we observed for APOBEC mRNA levels with MM patient outcomes. The MM microenvironment is characterized by a desynchronized cytokine milieu, with imbalanced pro- and anti-inflammatory factors that impact on MM and niche cells. Besides their general role in inflammatory processes, IL-11 as well as TGF-ß have both been implicated in the growth and differentiation block of osteoblasts [14], which in turn modulates MM cell activity. Likewise, APOBEC3B and APOBEC2 upregulation has been linked to systemic inflammation [13], suggesting that a pro-inflammatory microenvironment in MM cells could drive APOBEC2 and APOBEC3B expression. However, the precise regulation and function of APOBEC2 and APOBEC3B in MM cells still needs to be defined.

Table 1 Incorporation of EI-score gene expression information improves the performance of established risk classifiers in the MMRF CoMMpass dataset.
Full size table
Fig. 1: The EI-score reclassifies MM patients and identifies novel prognostic MM subgroups.
figure 1

Shown are graphical representations of OS Kaplan–Meier estimates based on the application of the EI-score[OS] to (A) MMRF CoMMpass patients who were stratified into ISS and R-ISS stage II and III as well as into R2-ISS low intermediate, high intermediate, and high risk groups. B MMRF CoMMpass patients carrying del(17p), t(4;14), or +1q, and (C) IFM/DFCI patients carrying del(17p), t(4;14), or +1q reclassified by the EI-score.

Full size image

In this study, we have developed the EI-score which serves as an important proof-of-concept, demonstrating that inclusion of molecular markers that reflect disease progression can improve MM risk assessment. Although our data highlights the limitations of cytogenetics-based risk stratifiers, ISS, R-ISS and R2-ISS represent the current clinical standard due to their accessibility. Eventually, the development of more contemporary stratification systems will be necessary to improve risk- and treatment stratifications of MM patients.